This document explains the instructions to run the DeepDIG project
-
Clone the project by running
git clone git@github.com:hamidkarimi/DeepDIG.git
in a path on your machine. Let's call this path CodePath -
Run
initial_script.py
. It prompts for a path on your machine to create a directory holding the data, results, etc. Make sure there is enough space in that path where you place the project (at least a couple of GBs). Let's call this path ProjectPath -
Copy or cut Data from the cloned Github repository to ProjectPath/DeepDIG/
-
MNIST and FASHIONMNIST data are already uploaded. For CIFAR10, download the data from here, unzip it and copy the files into ProjectPath/DeepDIG/CIFAR10
-
In
config.py
change the variable PATH to ProjectPath i.e., the path that you entered when runninginitial_script.py
-
Open a terminal and go to the upper-level directory containing the DeepDIG code i.e., where you cloned the code
/cd _CodePath
-
run
python -m DeepDIG.PreTrainedModels.{DATASET}.{MODEL}.train --dataset {DATASET} --pre-trained-model {MODEL}
whereDATASET
is the name of the dataset andMODEL
is the name of the modelExample:
python -m DeepDIGCode.PreTrainedModels.FASHIONMNIST.CNN.train --dataset FASHIONMNIST --pre-trained-model CNN
this will train the CNN model for FASHIONMNIST and then saves it.
Please refer to here to see how you can run DeepDIG against your new dataset/model.
Running the DeepDIG framework (Figure 2)
- Open a terminal and go to the upper-level directory containing DeepDIG code where you cloned the code
/cd _CodePath
- Run
python -m DeepDIGCode.main --dataset {DATASET} --pre-trained-model {MODEL} --classes {s};{t}
whereDATASET
is the name of the dataset,MODEL
is the name of the model, and s and t are two classes in the dataset for which you intend to DeepDIG
Example 2: python -m DeepDIGCode.main --dataset FASHIONMNIST --pre-trained-model CNN --classes 1;2
this will run DeepDIG against the trained CNN on FASHIONMNIST to characterize the decision boundary of classes 1 and 2 (i.e., Trouser and Pullover)
Note 1. See here for the explantion of DeepDIG's arguments.
Note 2. That model should be trained before as explained above.
Note 3. Classes are referred numerically from 0 to n-1 where n is the number of classes. For instance, you can find the classes of CIFAR10 here. See the following examples
Example 2: python -m DeepDIGCode.main --dataset CIFAR10 --pre-trained-model GoogleNet --classes 1;2
(automobile, bird)
Example 3: python -m DeepDIGCode.main --dataset CIFAR10 --pre-trained-model ResNet --classes 4;8
(deer, ship)
- All results including visualizations will be saved in the _CodePath/DeepDIG/{DATASET}/{PretrainedModel}/{(s,t)} where DATASET is the input dataset, PretrainedModel is the given pre-trained model, and s and t are input classes e.g. _CodePath/DeepDIG/CIFAR10/ResNet/(4,8) in Example 3
If you use the code in this repository, please cite the following papers
@article{karimi2019characterizing, title={Characterizing the Decision Boundary of Deep Neural Networks}, author={Karimi, Hamid and Derr, Tyler and Tang, Jiliang}, journal={arXiv preprint arXiv:1912.11460}, year={2019} }
@inproceedings{karimi2020decision, title={Decision Boundary of Deep Neural Networks: Challenges and Opportunities}, author={Karimi, Hamid and Tang, Jiliang}, booktitle={Proceedings of the 13th International Conference on Web Search and Data Mining}, pages={919--920}, year={2020} }
Web page: http://cse.msu.edu/~karimiha/ Email: karimiha@msu.edu